https://gcc.gnu.org/bugzilla/show_bug.cgi?id=128180
--- Comment #2 from Konstantinos Eleftheriou <konstantinos.eleftheriou at vrull dot eu> --- We have submitted a patch for this: https://gcc.gnu.org/pipermail/gcc-patches/2026-October/735814.html. The approach has been rejected, I'm putting it here mostly for the conversation history. With the patch, reduced.c and the two exp2f loops in LogOpCPU.cpp (AntiLogRenderer and Log2LinRenderer) are vectorized, and the latter call _ZGVnN4v_exp2f. The two log2f loops (LogRenderer and Lin2LogRenderer) are still rejected, for a separate reason: log2 (max (FLT_MIN, x)) becomes a branch between log2f (x) and the constant -126.0, so if-conversion turns the call into .MASK_CALL (__builtin_log2f, ...), and the vectorizer reports "relevant stmt not supported", since glibc declares the vector log2f as notinbranch only.
